Lane-based probabilistic motion prediction of surrounding vehicles and predictive longitudinal control method and apparatus

Abstract

Disclosed are probabilistic prediction for a motion of a lane-based surrounding vehicle and a longitudinal control method and apparatus using the same. The method includes obtaining surrounding vehicle information using a sensor, predicting a target lane of the surrounding vehicle based on the obtained surrounding vehicle information, performing future driving trajectory prediction for each target lane based on the surrounding vehicle information, and computing a probability of a collision likelihood based on a target lane and trajectory predictions of the surrounding vehicle in which future uncertainty has been taken into consideration and performing longitudinal control for collision avoidance.

What is claimed is: 
     
       1. A method of predicting a motion of surrounding vehicles and controlling an ego vehicle, the method comprising:
 obtaining surrounding vehicle information using a sensor; 
 predicting a target lane of the surrounding vehicle from a plurality of potential target lanes based on the obtained surrounding vehicle information, the plurality of potential target lanes being laterally adjacent to one another; 
 performing future driving trajectory prediction for each of the plurality of potential target lanes including the predicted target lane based on the surrounding vehicle information; and 
 computing a probability of a collision likelihood based on the target lane and trajectory predictions of the surrounding vehicle in which future uncertainty has been taken into consideration and performing longitudinal control for collision avoidance; 
 wherein predicting the target lane of the surrounding vehicle includes using an artificial neural network structure in which current and past-time series position information of the surrounding vehicle and road information for a predetermined time are used as input values. 
 
     
     
       2. The method of  claim 1 , wherein predicting the target lane of the surrounding vehicle comprises:
 outputting a probability of the target lane as an output value. 
 
     
     
       3. The method of  claim 2 , wherein predicting the target lane of the surrounding vehicle is performed using Interacting Multiple Model, or Markov Chain in addition to an artificial neural network structure. 
     
     
       4. The method of  claim 1 , wherein performing the driving trajectory prediction comprises:
 using an artificial neural network structure in which current and past time-series longitudinal/lateral positions and/or velocity of the surrounding vehicle for a predetermined time are used as input values, and 
 outputting longitudinal/lateral positions as an output value. 
 
     
     
       5. The method of  claim 4 , wherein performing the driving trajectory prediction is performed using Polynomial fitting using the artificial neural network position outputs. 
     
     
       6. The method of  claim 4 , wherein a probability of the longitudinal/lateral positions is output as a final output value through a process of transforming the deterministic prediction output values, output as the longitudinal/lateral positions, into a probabilistic prediction output values. 
     
     
       7. The method of  claim 1 , wherein computing the probability of the collision likelihood and performing the longitudinal control comprises:
 performing optimal longitudinal control that minimizes a cost function, a difference between a longitudinal target velocity and a current velocity and target acceleration, while a collision probability does not exceed a predetermined value. 
 
     
     
       8. The method of  claim 7 , wherein:
 longitudinal control in which uncertainty for a collision has been taken into consideration is performed using longitudinal safety distance restriction between an ego vehicle and the surrounding vehicle in a chance-constraint form through probabilistic motion prediction for a motion of the surrounding vehicle, and 
 a desired driving style of the automated driving algorithm is adjusted by controlling a chance-constraint parameter. 
 
     
     
       9. An apparatus for predicting a motion of a surrounding vehicle and controlling an ego vehicle, the apparatus comprising:
 a sensor configured to obtain surrounding vehicle information using a sensor; 
 a prediction unit configured to predict a target lane of the surrounding vehicle from a plurality of potential target lanes based on the obtained surrounding vehicle information, the plurality of potential target lanes being laterally adjacent to one another; 
 a probability calculation unit configured to perform future driving trajectory prediction for each of the plurality of potential target lanes including the predicted target lane based on the surrounding vehicle information; and 
 a longitudinal controller configured to compute a probability of a collision likelihood based on the target lane and trajectory predictions of the surrounding vehicle in which future uncertainty has been taken into consideration and perform longitudinal control for collision avoidance; 
 wherein the prediction unit is configured to use an artificial neural network structure in which current and past time-series position information and road information of the surrounding vehicle for a predetermined time are used as input values. 
 
     
     
       10. The apparatus of  claim 9 , wherein the prediction unit is configured to:
 output a probability of the target lane as an output value. 
 
     
     
       11. The apparatus of  claim 9 , wherein the probability calculation unit is configured to:
 use an artificial neural network structure in which current and past time-series longitudinal/lateral positions and/or velocity of the surrounding vehicle for a predetermined time are used as input values, and 
 output longitudinal/lateral positions locations as an output value. 
 
     
     
       12. The apparatus of  claim 9 , wherein the longitudinal controller is configured to:
 perform optimal longitudinal control using a cost function to minimize a difference between a longitudinal target velocity and a current velocity and target acceleration so that a collision probability does not exceed a predetermined value and the cost function is minimized. 
 
     
     
       13. A longitudinal control method using lane-based probabilistic prediction for a motion of a surrounding vehicle, the method comprising:
 obtaining surrounding vehicle information using a sensor; 
 predicting a target lane of the surrounding vehicle from a plurality of potential target lanes based on the obtained surrounding vehicle information using an artificial neural network structure, the plurality of potential target lanes being laterally adjacent to one another; 
 performing future driving trajectory prediction for each of the plurality of potential target lanes including the predicted target lane based on the obtained surrounding vehicle information; and 
 computing a cost function to minimize a difference between a longitudinal target velocity and a current velocity and target acceleration using the predicted target lane and driving trajectory predictions of the surrounding vehicle and performing optimal longitudinal control so that a collision probability does not exceed a predetermined value and the cost function is minimized. 
 
     
     
       14. The method of  claim 13 , wherein:
 longitudinal control in which uncertainty for a collision has been taken into consideration is performed using longitudinal safety distance restriction between an ego vehicle and the surrounding vehicle in a chance-constraint form through probabilistic motion prediction for a motion of the surrounding vehicle, and 
 a desired driving style of the automated driving algorithm is adjusted by controlling a chance-constraint parameter. 
 
     
     
       15. A longitudinal control apparatus, comprising:
 a sensor configured to obtain surrounding vehicle information using a sensor; 
 a prediction unit configured to predict a target lane of the surrounding vehicle from a plurality of potential target lanes based on the obtained surrounding vehicle information using an artificial neural network structure, the plurality of potential target lanes being laterally adjacent to one another; 
 a probability calculation unit configured to perform future driving trajectory prediction for each of the plurality of potential target lanes including the predicted target lane based on the obtained surrounding vehicle information; and 
 a predicted target lane and driving trajectory predictions of the surrounding a longitudinal controller configured to compute a cost function to minimize a difference between a longitudinal target velocity and a current velocity and target acceleration using the predicted target lane and driving trajectory predictions of the surrounding vehicle and perform optimal longitudinal control so that a collision probability does not exceed a predetermined value and the cost function is minimized. 
 
     
     
       16. The apparatus of  claim 15 , wherein:
 longitudinal control in which uncertainty for a collision has been taken into consideration is performed using longitudinal safety distance restriction between an ego vehicle and the surrounding vehicle in a chance-constraint form through probabilistic motion prediction for a motion of the surrounding vehicle, and 
 a desired driving style of the automated driving algorithm is adjusted by controlling a chance-constraint parameter.